OpenAI Last week saw the official launch of the much-anticipated search product ChatGPT Search, launched a direct challenge to Google. The arrival of this moment, in fact, has long been expected in the industry, so Google earlier this year will be generated by artificial intelligence answers into its core products, but in the process there are some embarrassing "illusion". This mistake has led many to believe that OpenAI'sSearch EnginesIt's going to be a real "Google killer".
However, after using ChatGPT Search as the default search engine for a day or so, tech journalist Maxwell Zeff quickly switched back to Google. While OpenAI's search product is certainly impressive in some ways, and gives us a glimpse of what the future of AI search interfaces might look like, at this stage it's still a bit too idealized to be used as an everyday search tool.
ChatGPT Search does provide answers to real-time questions efficiently at times, avoiding the hassle of flipping through advertisements and SEO-optimized articles. Ultimately, it also gives succinct answers in a clear and concise format with a link to the source of the information, the title, and a short introduction on the right side so that we can confirm that the AI-generated text we're reading is accurate.
However,ChatGPT Search often feels impractical for everyday search needs.
In its current form, ChatGPT Search is not reliable enough to handle the short queries that people use Google for most on a daily basis. These types of queries are usually no more than four words, yet they make up the majority of Google searches; usually a few keywords that are used to quickly find the desired web page. Many people may not realize that they perform such searches all day long, and Google has been performing quite well in this area.
For example, queries such as "NBA Celtics scores," "cotton socks," "library hours," "weather in San Francisco," "cafes near me," etc. have made Google the preferred access point to the Internet for billions of people. "Cotton socks," "library hours," "weather in San Francisco," "cafes near me," and so on, are the kinds of queries that have made Google the preferred gateway to the Internet for billions of people.
Zev was often frustrated while testing with ChatGPT Search, which made him realize that he was doing a lot of keyword searches every day.He was unable to consistently find the information he needed on ChatGPT Search with a short query.. For the first time in years, Zev was genuinely eager to get back to using Google search.
Of course, that's not to say that Google's search quality hasn't declined over the past decade; in fact, it has indeed declined due to advertising and search engine optimization. However, Zev had to keep Google open in another window at all times during the test because ChatGPT Search often failed to provide him with the correct answers or links to web pages.
Who would win: a ChatGPT Search or a short query?
Zev tried to view the Denver Nuggets-Minnesota Timberwolves NBA game in real time by typing in "Nuggets Score". However, ChatGPT replied that the Nuggets were winning, when in fact they were losing. Even worse, according to Google's results, the Timberwolves score shown by ChatGPT is a full 10 points lower than the actual score.
In another instance, Zev typed in "today's earnings" to look up quarterly company reports that could affect Friday's stock price, and ChatGPT told him that Apple and Amazon would be reporting on Friday, when in fact both companies had already reported a day earlier. In a nutshell.ChatGPT is "hallucinating" and providing incorrect information.
In another test, Zev typed in the name of a tech executive hoping to find his contact information, and ChatGPT not only showed him a summary of the executive's Facebook profile, but also "made up" a link to his LinkedIn page. But when Zev clicked on the link, he received an error message.
On another occasion, Zev typed in "baggy jeans" in the hopes of finding a source to buy them, and ChatGPT Search first explained what baggy jeans were (although the definition is probably redundant for most people), and then suggested that he go to Amazon.com to buy a nice pair of jeans.
There are many more examples like this, but I'm sure you get the idea. On the first day of using ChatGPT Search, theZev encountered many invalid links, information "illusions" and random answers.
It may be a "Google killer" in the future, but not today.
The release of ChatGPT Search is a milestone for OpenAI. Sam Altman, the company's CEO, has often downplayed his own startup's AI capabilities, but he still praised ChatGPT Search, saying it was "really good". Perhaps the difference this time around is that search is one of the biggest businesses on the Internet, and OpenAI's search capabilities could indeed pose a substantial threat to its biggest competitor, Google.
While Google Search is, objectively speaking, a 25-year-old product, ChatGPT Search is completely new, and OpenAI said in a blog post that they plan to keep improving this feature in the coming months based on user feedback, which will undoubtedly be a key area of investment for the startup to focus on.
To be sure, ChatGPT Search does a pretty good job of answering complex and lengthy research questions. For example, Google may not be able to quickly provide a satisfactory answer to the question, "Which American professional sports league has the most diversity?" For example, Google may not be able to give you a quick and satisfactory answer to this question, but ChatGPT Search is able to quickly consolidate information from multiple sites and provide you with a fairly complete answer in a matter of seconds. Similarly, the Perplexity search product, which has been around for over a year, does a good job with these kinds of questions.
ChatGPT Search is now web-accessible compared to the legacy version of ChatGPT, providing users with a superior web browsing interface. ChatGPT will now provide clearer links to sources when accessing information -- for news stories, ChatGPT will work with major news media companies that already have all licensing agreements in place.
However, most Google search queries are not so lengthy. To truly replace Google, OpenAI needs to improve ChatGPT Search's performance in more useful, shorter everyday searches.
In response, OpenAI has not shied away from acknowledging the shortcomings of ChatGPT Search for short queries. With ChatGPT Search, we've observed that users tend to ask questions in a more natural, colloquial way than they have in the past with other search tools," said Niko Felix, a spokesperson for the company. At the same time, we've also noticed that queries with short web navigation -- often just a few words -- are also very prevalent. We plan to keep optimizing to improve the user experience for these types of queries in the future."
In other words, these short keyword queries are where Google has long been an integral part of its strengths, and until OpenAI is able to perfectly address these types of queries, Google will remain the search tool of choice for many.
OpenAI may face additional challenges in processing these short queries for several reasons: first, ChatGPT is technically dependent on Microsoft's Bing search engine, which is widely regarded in the industry as being inferior to Google in terms of search quality. Second, large language models may not be very good at handling these short inputs. Such models usually need to receive complete written prompts in order to produce valid answers. Perhaps we need to redirect users in some way -- using large language models to convert short queries into more detailed questions before ChatGPT Search can excel at these types of searches.
It's worth noting that while OpenAI's search product has just been released, another AI search tool, Perplexity, which has also been labeled a "Google killer," is already processing 100 million search queries per week, but it's having the same problem with short queries.
Aravind Srinivas, CEO of Perplexity, has discussed the differences in how people use his product versus a Google search. He says, "The average query in Google search is usually between 2 and 3 words, whereas in Perplexity the average query is 10 to 11 words. This suggests that on Perplexity, people are more likely to ask direct questions. In contrast, on Google, users simply type in a few keywords and are quickly redirected to relevant links."
Zev argues that the fact that people aren't using these new products for web navigation actually reveals a more serious problem than OpenAI or Perplexity have publicly acknowledged. This means that ChatGPT Search and Perplexity are not replacing Google Search's core strengths in web navigation. Instead, these AI products are carving out a new niche, revealing and presenting information that has been overlooked or hidden in traditional search. This, of course, is a highly valuable area in its own right.
Both OpenAI and Perplexity say they are confident that they can improve the processing of these short queries in the future. But until then, Zev doesn't see either product as a complete replacement for Google. If OpenAI really wants to be the new gateway to the Internet, it'll have to create a new portal that's even better and more convenient than the existing one.